结合时变滤波和时频掩码的语音增强方法

本文提出了一种结合鲁棒时变滤波和时频掩码的语音增强方法。首先在带噪语音的时频域中,结合图像处理方法估计出初始瞬时频率信息。然后基于该瞬时频率信息,利用鲁棒时变滤波算法构建降噪后的语音信号。最后根据重构语音的时频特征预测时频掩码。该掩码在带噪语音的时频域中能够有效地保留语音成分且抑制噪声成分,从而达到语音增强的目的。实验结果表明,在几种常见背景噪声环境下,所提语音增强算法在抑制背景噪声干扰、提升语音整体质量方面表现良好,尤其是在低信噪比环境下具有明显的优势。

Joint Time-Varying Filtering and Masking for Speech Enhancement

In this paper, a speech enhancement method combining the robust time-varying filtering algorithm with time-frequency mask is proposed. Firstly, in the time frequency distribution of the noisy speech signal, a series of image processing methods are exploited to acquire the initial instantaneous frequency information of the signal. Next, based on the instantaneous frequency information, the reconstructed speech signal with less noise is obtained via the robust time-varying filtering algorithm. Finally, we predict the time-frequency mask by utilizing the energy distribution of the reconstructed signal. The mask can efficiently keep speech information and suppress the noise in the time-frequency domain, which is used to enhance the speech signal. The proposed method is evaluated and compared with some conventional speech enhancement methods, and results in terms of different evaluation metrics demonstrate that the proposed speech enhancement method is effective in suppressing background noise and improving the overall quality of speech, especially in low SNR environments.